If you can read this, either the style sheet didn't load or you have an older browser that doesn't support style sheets. Try clearing your browser cache and refreshing the page.

(PC Magazine)   Own a computer with an Intel chip? Be ready for a 30% boost in performance. Wait, did I say boost? I meant reduction   ( uk.pcmag.com) divider line
    More: Fail, design flaw, Operating system, major design flaw, kernel memory, Intel, chip design flaw, X86, Intel Corporation  
•       •       •

4448 clicks; posted to Geek » on 03 Jan 2018 at 9:20 AM (27 weeks ago)   |   Favorite    |   share:  Share on Twitter share via Email Share on Facebook   more»



207 Comments     (+0 »)
 
View Voting Results: Smartest and Funniest


Oldest | « | 1 | 2 | 3 | 4 | 5 | » | Newest | Show all

 
2018-01-03 07:22:14 AM  
 
2018-01-03 07:29:58 AM  
argh. There goes my new year.
 
2018-01-03 08:31:47 AM  
Its gotta be a flaw in one of the instruction set extensions

The question is which one
 
2018-01-03 08:32:03 AM  
Can I hear a Class-Action lawsuit amen?
 
2018-01-03 08:40:33 AM  
<tinfoil>
Perfect cover for ISP throttling shenanigans
</tinfoil>
 
2018-01-03 08:41:00 AM  
PC MASTER RACE CONFIRMED
 
2018-01-03 08:43:58 AM  
The 30% is probably worst case scenario. I'm waiting for real-world results before I break out the outrage gifs.
 
2018-01-03 08:46:22 AM  
Relax guys, they're just throttling the CPU to preserve your laptop's battery.
 
2018-01-03 08:52:33 AM  
My 30 year preference for AMD pays off!

[zoidberg.gif]
 
2018-01-03 08:59:47 AM  
The kernel is the core of an operating system which controls anything and everything running on a system.

That's what I put on my Operating Systems final in college.
 
ZAZ [TotalFark]
2018-01-03 09:14:50 AM  
Based on scanning the internets over the past couple days, it appears to me the fix isn't too costly on recent chips which don't flush the TLB as often. The cost also depends heavily on the system call rate of your program. Computation will fare better than filesystem traversal. No idea what video games are like as a workload. They do a lot of I/O to the graphics card, but maybe not so many system calls.
 
2018-01-03 09:16:06 AM  
What did Apple do differently that their Intel chips don't have this issue? or does PC mag just pretend Apple doesn't exist...
 
2018-01-03 09:25:06 AM  

sno man: What did Apple do differently that their Intel chips don't have this issue? or does PC mag just pretend Apple doesn't exist...


They said linux, windows, and MacOS in the article
 
2018-01-03 09:27:42 AM  
Still no cure for the bug ridden performance sucking gaping security hole that is IME?
 
2018-01-03 09:29:14 AM  

sno man: What did Apple do differently that their Intel chips don't have this issue? or does PC mag just pretend Apple doesn't exist...


Right at the top of the page "Patches for Linux, Windows, and macOS are required, with the side effect potentially being a major system performance hit", and then again in the article, right in the second paragraph "Those patches will have to be made for Linux, Windows, and macOS. "
 
2018-01-03 09:32:50 AM  
My 7 year old AMD chip is still chugging along and doing just fine.
 
2018-01-03 09:34:12 AM  
I have an intel i5 that i bought around December 2009 or so. Am I boned?
 
2018-01-03 09:35:23 AM  

Mercutio879: The 30% is probably worst case scenario. I'm waiting for real-world results before I break out the outrage gifs.


Even the 5% cited as low-end is huge if you're doing any kind of significant CPU over-commit on your hypervisor hosts.
 
2018-01-03 09:36:16 AM  

OldJames: sno man: What did Apple do differently that their Intel chips don't have this issue? or does PC mag just pretend Apple doesn't exist...

They said linux, windows, and MacOS in the article

Snuffybud: sno man: What did Apple do differently that their Intel chips don't have this issue? or does PC mag just pretend Apple doesn't exist...

Right at the top of the page "Patches for Linux, Windows, and macOS are required, with the side effect potentially being a major system performance hit", and then again in the article, right in the second paragraph "Those patches will have to be made for Linux, Windows, and macOS. "


this is why PBC posting before coffee is a bad idea kids.
 
2018-01-03 09:37:39 AM  
Me too. Yay AMD
 
2018-01-03 09:37:49 AM  

sno man: What did Apple do differently that their Intel chips don't have this issue? or does PC mag just pretend Apple doesn't exist...


No...we have this issue too..
 
2018-01-03 09:42:23 AM  

Carousel Beast: Mercutio879: The 30% is probably worst case scenario. I'm waiting for real-world results before I break out the outrage gifs.

Even the 5% cited as low-end is huge if you're doing any kind of significant CPU over-commit on your hypervisor hosts.


Not significantly overcommitted. My constraints have always been memory, rather than CPU. We're a fairly small shop, though, we only have 9 blades with ~300 VMs, most of those being VDI.

This will probably make us look at AMD based servers a little more closely in the future, as long as I can vMotion to them. I know VMware isn't very keen on having AMD and Intel servers in the same cluster.
 
2018-01-03 09:44:15 AM  
 
2018-01-03 09:45:12 AM  
from a thread over at Ars:
"Ruh-oh, looks like Intel's CEO sold enough stock to only have the bare minimum required for his position:
Intel's CEO Just Sold a Lot of Stock "

Interesting if this really has been percolating since November...
 
2018-01-03 09:47:26 AM  

AntonChigger: I have an intel i5 that i bought around December 2009 or so. Am I boned?


Wondering the same thing about my dual-Westmere Xeon box I'm farking on right now.
 
2018-01-03 09:49:15 AM  

GardenWeasel: <tinfoil>
Perfect cover for ISP throttling shenanigans
</tinfoil>


I was thinking more along the lines of no one is buying computers anymore, as their ten year old laptop still does everything the majority of people need it to. Hell even professional are getting by with their five year old machines. Gamers are the only ones I know that always want the newest, but most of them I know upgrade their video cards far more often than processors.

So if you slow everything down they'll want to buy a new one.

/tinfoil hat aside, next computer I get I'm going back to AMD.
 
2018-01-03 09:50:32 AM  

BMFPitt: My 7 year old AMD chip is still chugging along and doing just fine.


Glad your investment has finally paid off. :-)
 
2018-01-03 09:52:58 AM  

cman: Its gotta be a flaw in one of the instruction set extensions

The question is which one


Follow the link to the Register article for a more technical explanation. Long story short, the processor is executing speculative commands without security.
 
2018-01-03 09:54:44 AM  

Mercutio879: Carousel Beast: Mercutio879: The 30% is probably worst case scenario. I'm waiting for real-world results before I break out the outrage gifs.

Even the 5% cited as low-end is huge if you're doing any kind of significant CPU over-commit on your hypervisor hosts.

Not significantly overcommitted. My constraints have always been memory, rather than CPU. We're a fairly small shop, though, we only have 9 blades with ~300 VMs, most of those being VDI.

This will probably make us look at AMD based servers a little more closely in the future, as long as I can vMotion to them. I know VMware isn't very keen on having AMD and Intel servers in the same cluster.


My constraint is CPU; specifically the per core licensing of sql. I'm overcommitted about 2:1 now and am on a 3 year plan to get down to 1.25:1, but if this actually has teeth as a performance hit, I may commandeer the Wintel budget to buy new hosts.
 
2018-01-03 09:58:39 AM  

hashtag.acronym: AntonChigger: I have an intel i5 that i bought around December 2009 or so. Am I boned?

Wondering the same thing about my dual-Westmere Xeon box I'm farking on right now.


I read somewhere it goes back as far as Pentium III.
 
2018-01-03 09:58:50 AM  
The article I read last night said while AMD chips don't have the problem, the Linux and presumed Windows fixes apply the penalty for both. Perhaps an update will follow to be more selective, and they just need to get a reliable fix out sooner than that would allow? Still plenty for AMD and fans to crow about, but much more if they can escape the Intel penalty. AMD might even implement or bankroll such a fix for Linux.
 
2018-01-03 10:01:29 AM  
Ugh.  My work laptop is already bare bones as it is.

Less worried about the custom rig as I went overboard on certain specs when I made it, but processors are expensive and it was built two years ago.  :( Booo.
 
2018-01-03 10:04:57 AM  
It's not going to be a 30% reduction across the board, it's a 30% reduction on specific tasks. If you happen to do these specific tasks all the time then yes you're boned. If you don't your actual performance hit is going to be better.

Looking at how Intel has handled these issues in the past, the F00F bug was able to be resolved with little performance penalty (extra page table traffic, and only when dealing with the bug) and the solution only required moving some kernel data structures around in memory. The FDIV bug required software to workaround, which introduced a big performance penalty, but Intel recalled and offered to replace those chips that were affected (eventually).
 
2018-01-03 10:06:21 AM  
The performance hit will be task-related.

Specifically calls to the OS, shelling out to Kernel, that will be effected more than anything. So your game-loop won't be bad, but business software that opens Outlook, or Excel automation (you know real stuff, like I make) will grind to a halt.

/and all your could apps are screwed too.
//I don't write those so, ha ha
/// and three
 
ZAZ [TotalFark]
2018-01-03 10:09:22 AM  
pdieten

It's not clear yet if the bug allows direct user mode access to kernel memory, or if it another timing bug that depends on behavior of instructions that are aborted before completion. The first would mean the chip does not work to spec, and Intel should be flogged mercilessly. The second is the kind of bug people used to not think about because faster was always better.

The Register article links to somebody who tried the second kind of attack and didn't get it to work, but maybe didn't try hard enough.

Hypothetical example of the first kind of bug: A load instruction issued right before a system call is aborted due to an L1 cache miss and is put back on the load queue, but mistakenly given the new processor privilege level.

Real example of the second kind of bug: a prefetch instruction takes a different amount of time to execute depending on whether the address is valid and unreadble or not valid at all, so it can be used to probe the validity of unreadable kernel addresses.
 
2018-01-03 10:13:36 AM  

Mike_LowELL: PC MASTER RACE CONFIRMED


Ja, ja, jaaaa. We all know about M$.
 
2018-01-03 10:13:52 AM  

Raider_dad: Me too. Yay AMD


I'm building a new box But haven't gotten around to getting my CPU yet. Does this mean the price of Ryzen 7's are gonna go up?
 
2018-01-03 10:15:39 AM  

wrenchboy: Raider_dad: Me too. Yay AMD

I'm building a new box But haven't gotten around to getting my CPU yet. Does this mean the price of Ryzen 7's are gonna go up?


Tough to say, but with Ryzen 2 expected to launch in March, it may be a good idea to hold off if you can wait.
 
2018-01-03 10:16:26 AM  

tjsands1118: GardenWeasel: <tinfoil>
Perfect cover for ISP throttling shenanigans
</tinfoil>

I was thinking more along the lines of no one is buying computers anymore, as their ten year old laptop still does everything the majority of people need it to. Hell even professional are getting by with their five year old machines. Gamers are the only ones I know that always want the newest, but most of them I know upgrade their video cards far more often than processors.

So if you slow everything down they'll want to buy a new one.

/tinfoil hat aside, next computer I get I'm going back to AMD.


Difficulty: paying out class action lawsuits and making other customer reparations for the fark up tends to wipe out any extra profit you would have made off of driving a faster upgrade cycle.
 
2018-01-03 10:16:40 AM  
img.fark.netView Full Size
 
2018-01-03 10:17:15 AM  

pdieten: cman: Its gotta be a flaw in one of the instruction set extensions

The question is which one

Follow the link to the Register article for a more technical explanation. Long story short, the processor is executing speculative commands without security.


I was right with them all the way through speculative execution and security checks but then they started talking about boffins and I was like WTF
 
2018-01-03 10:17:43 AM  
Would macOS, being based on a microkernel, be any more or less vulnerable to this than NT or Linux?
 
2018-01-03 10:18:10 AM  

wrenchboy: Does this mean the price of Ryzen 7's are gonna go up?


Relative to the price of an Intel chip of the same speed the price will drop.
 
2018-01-03 10:19:01 AM  

Carousel Beast: Mercutio879: Carousel Beast: Mercutio879: The 30% is probably worst case scenario. I'm waiting for real-world results before I break out the outrage gifs.

Even the 5% cited as low-end is huge if you're doing any kind of significant CPU over-commit on your hypervisor hosts.

Not significantly overcommitted. My constraints have always been memory, rather than CPU. We're a fairly small shop, though, we only have 9 blades with ~300 VMs, most of those being VDI.

This will probably make us look at AMD based servers a little more closely in the future, as long as I can vMotion to them. I know VMware isn't very keen on having AMD and Intel servers in the same cluster.

My constraint is CPU; specifically the per core licensing of sql. I'm overcommitted about 2:1 now and am on a 3 year plan to get down to 1.25:1, but if this actually has teeth as a performance hit, I may commandeer the Wintel budget to buy new hosts.


This raises an interesting question for me:  Are hypervisors going to be double-dinged?  Like if the hypervisor applies the more costly work-around in itself...and then the VMs' OSes also do the same are the VMs going to be effectively doubling-down the performance?

Hrm.  When I read about this last night it did cause me to go look at some of those AMD Eypx chip benchmarks and I gotta say they do look pretty good...even without this revelation regarding Intel's stuff in play.
 
2018-01-03 10:19:30 AM  

triptheory: https://www.fool.com/investing/2017/1​2​/19/intels-ceo-just-sold-a-lot-of-stoc​k.aspx

I'll just leave this here...


Oooooh, sounds like somebody is going to have a chat with the SEC soon...
 
2018-01-03 10:20:33 AM  

pheelix: I read somewhere it goes back as far as Pentium III.


"I am Pentium of Borg. Division is futile, you will be approximated!"

I'm wondering if they found that "Row Hammer" was more exploitable than previously thought

https://en.wikipedia.org/wiki/Row_ham​m​er

I always considered that if your memory is pattern sensitive at all, you are doing it wrong.

/Intel should go F00F themselves
 
2018-01-03 10:22:07 AM  

sno man: What did Apple do differently that their Intel chips don't have this issue? or does PC mag just pretend Apple doesn't exist...


FTFA:Patches for Linux, Windows, and macOS are required, with the side effect potentially being a major system performance hit.
 
2018-01-03 10:22:49 AM  
the other side of this coin is that the only real fix for this will be to throw away all of your computers and buy ones w/ post-2017 CPUs.

And of course Win 7 can't tell this and will have to have the have the PC killing "fix" whether or not they have a vulnerable CPU.  While Win10 can be selective about it. So they'll make this another nail in the coffin of Win7.

/and all those XP machines still out there (in ATMs and at Foxconn) won't be patched at all.
 
2018-01-03 10:26:13 AM  

Mike_LowELL: PC MASTER RACE CONFIRMED


Games are unaffected.

/also, I heard this weird rumour, that PCs doesn't have to, *GASP*, run Intel CPUs!
 
2018-01-03 10:28:01 AM  

enry: My 30 year preference for AMD pays off!

[zoidberg.gif]


I'm hoping my 6 month preference for AMD stock will pay off now as well.

img.fark.netView Full Size
 
Displayed 50 of 207 comments


Oldest | « | 1 | 2 | 3 | 4 | 5 | » | Newest | Show all


View Voting Results: Smartest and Funniest

This thread is closed to new comments.

Continue Farking

On Twitter





Top Commented
Javascript is required to view headlines in widget.
  1. Links are submitted by members of the Fark community.

  2. When community members submit a link, they also write a custom headline for the story.

  3. Other Farkers comment on the links. This is the number of comments. Click here to read them.

  4. Click here to submit a link.

Report